Key points are not available for this paper at this time.
Early prediction of in-hospital mortality in Intensive Care Unit (ICU) patients is crucial for optimizing resource allocation, informing prognosis discussions, and guiding treat-ment decisions. While conventional systems exist, they often lack precision and struggle to handle complex, multidimensional data with limited resource availability. This study investigates the potential use of machine learning (ML) techniques to enhance in-hospital mortality prediction in ICU patients with heart failure (HF). An ML-based approach has been proposed in this study utilizing electronic health records of ICU patients containing demographic, physiological, laboratory, and medication data. The significance of the features in the dataset has also been evaluated through a voting ensemble technique aggregating the results from multiple feature selection techniques. Various ML classification algorithms are trained, tested and compared to identify the model with the best predictive performance both with reduced feature set and all the features from the dataset. The model's generalizability and efficacy is evaluated through multiple performance parameters as well as error analysis. The Random Forest Classification model significantly outperforms other models attaining 96.8% accuracy & 0.032 Mean Absolute Error in predicting in-hospital mortality, achieving a higher area under the receiver operating characteristic curve (AUROC). Feature importance analysis here also reveals the most dominant 27 crucial attributes influencing mortality risk for this dataset, providing valuable insights for clinical decision-making. Thus, this study can lead to more efficient resource utilization, personalized treatment strategies, and ultimately, improved patient outcomes particularly for ICU patients.
Ali et al. (Thu,) studied this question.